HeyGen has launched Avatar 6, its newest AI model for talking avatars, which reads a script’s context and delivery and generates gestures to match. The model aims to end the stiff, repetitive motion that has made AI presenters look unnatural on screen.
Gestures that follow the script
Instead of looping the same movements, Avatar 6 times gestures to the words being spoken. A point, a count or a shift in expression arrives with the line it belongs to, so the avatar moves the way a person would while making the same point.
The model learns how each person actually moves from their own footage. An avatar can be built from as little as 30 seconds of video, topped up with clips from YouTube, Instagram or TikTok, so movement reflects the individual’s real habits, not a generic template.
Under the hood, the model’s runtime context window grows tenfold, from about three seconds to 30. That longer memory supports more varied movement, better lip sync and native 2K video output, while existing Personal Avatars upgrade automatically without new footage.
Two versions, one guardrail
Avatar 6 comes in two versions: the full model, built for the highest quality, and Avatar 6 Turbo, a faster and cheaper cut for quicker jobs. Both are available now on HeyGen’s platform and through its API, with support for select third-party platforms too.
Realistic videos of real people stay inside HeyGen’s own platform, where live identity verification and explicit consent are required. Third-party platforms may generate photo-based and virtual avatars only, a line HeyGen draws as AI likenesses grow harder to spot on screen.
Our Read
The realism numbers deserve a careful read. HeyGen says 84% of participants in its own Turing-style testing rated the avatar as equally realistic as the real person, and that it ranked first in New Measure’s Avatar Arena, but neither test has independent verification behind it.
Even so, the product step is real. Script-aware gestures attack the exact flaw users complain about most in AI avatars, and a tenfold context window is a concrete engineering gain rather than a marketing line, which is why the launch lands as a model release, not a rebrand.
HeyGen’s scale gives the launch weight. The company counts more than 40 million users across 196 countries, over 118 million videos created and adoption across 85% of the Fortune 100, and it passed $200 million in annual recurring revenue in June after doubling revenue in eight months.
Why It Matters
Talking-head video is now a working tool for businesses, not a novelty, and the uncanny gap is its biggest blocker. If an avatar’s gestures track the script, more companies will trust it with customer-facing video, which is the market HeyGen is built to serve.
Background
Avatar 6 follows HeyGen Video, the company’s first general-purpose video model, and HeyGen Voice, which debuted at number one on Artificial Analysis’ leaderboard. Together the three releases give HeyGen a proprietary stack across video, voice and avatars for content creation.
The video-avatar field keeps crowding. Krea Video already puts several rival generators in one browser workspace, and Kling’s funding round showed how much capital is chasing the same creators, so HeyGen’s answer is a better avatar, not a bigger catalogue of models.
Source: HeyGen on PR Newswire


0 Comments